The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill instructions direct users to download and install binary executables (excelcli.exe, mcp-excel.exe) and VS Code extensions from a non-trusted GitHub repository (github.com/sbroenne/mcp-server-excel).
[REMOTE_CODE_EXECUTION]: The installation process involves executing downloaded binaries on the host system to set up the MCP server and CLI tool, which is a high-risk operation when the source is unverified.
[COMMAND_EXECUTION]: The skill is built around a CLI wrapper (excelcli) that performs over 230 operations on the host machine via the Microsoft Excel COM API, including file management and window manipulation.
[DYNAMIC_EXECUTION]: The skill exposes tools to import and run VBA macros (vba import-module, vba run-macro) and create/execute Power Query M code. These are forms of scripting and code execution within the Excel environment that can be leveraged for malicious purposes.
[INDIRECT_PROMPT_INJECTION]: The skill's primary function is to ingest and process external data from Excel workbooks, CSV files, and VBA scripts. There are no mentioned boundary markers or sanitization procedures to prevent malicious instructions embedded in these files from influencing the agent's behavior.
Ingestion points: file open, powerquery create (File.Contents), vba import-module (source files), and range get-values from local workbooks.
Boundary markers: None detected in the instructions or command patterns.
Capability inventory: File system write access, VBA macro execution, Power Query execution, and screenshot capture.
Sanitization: No evidence of validation or sanitization for code or data imported into the Excel environment.
[DATA_EXFILTRATION]: The skill includes a screenshot tool capable of capturing ranges or entire sheets as PNG files. This provides a mechanism for an agent to capture sensitive visual data and potentially transmit it outside the local environment.

excel-mcp-automation